Digital Technology and the Future of Interstitial Lung Diseases 2 - Machine learning in radiology: the new frontier in interstitial lung diseases

Aug 4, 2026

Interstitial lung disease management faces major obstacles: early detection remains difficult, baseline data do not support accurate prognostication, and monitoring treatment response with high-resolution CT is limited by inter-reader variability and insufficient sensitivity over short follow-up. This Review summarizes clinical and research gaps in ILD diagnosis and prognosis and explains how machine learning can advance imaging biomarker research. Machine-learning models can detect ILD in at-risk groups, estimate fibrosis extent, link radiology to lung-function decline, and support endpoints in treatment trials. It highlights opportunities from deep learning and radiomics, and stresses algorithm consistency and biomarker validation against predictors of outcomes.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Challenges for ILD management

Clinical and research gaps in ILD

Early detection

Prognostication with baseline data

Monitoring response to therapy

Series Digital Technology and the Future of Interstitial Lung Diseases 2 Machine learning in radiology: the new frontier in interstitial lung diseases Hayley Barnes, Stephen M Humphries, Peter M George, Deborah Assayag, Ian Glaspole, John A Mackintosh, TameraJ Corte, Marilyn Glassberg, KerriA Johannson, Lucio Calandriello, Federico Felder, Athol Wells, Simon Walsh Challenges for the effective management of interstitial lung diseases (ILDs) include difficulties with the early detection of disease, accurate prognostication with baseline data, and accurate and precise response to therapy. The purpose of this Review is to describe the clinical and research gaps in the diagnosis and prognosis ofILD, and how machine learning can be applied to image biomarker research to close these gaps. Machine-learning algorithms can identify ILD in at-risk populations, predict the extent of lung fibrosis, correlate radiological abnormalities with lung function decline, and be used as endpoints in treatment trials, exemplifying how this technology can be used in care for people with ILD. Advances in image processing and analysis provide further opportunities to use machine learning that incorporates deep-learning-based image analysis and radiomics. Collaboration and consistency are required to develop optimal algorithms, and candidate radiological biomarkers should be validated against appropriate predictors of disease outcomes. Introduction The field of interstitial lung disease (ILD) is at acrossroads. Historically, assessment of ILD has focused on a diagnosis dependent on a complex clinical, radiological, and pathological process, for which radio pathological patterns are fundamental. However, these patterns do not reliably inform disease behaviour, particularly in patients who do not present with usual interstitial pneumonia in a surgical lung biopsy.1 Although idiopathic pulmonary fibrosis (IPF) is the prototypical progressive fibrotic lung disease, progressive disease behaviour is not limited to patients with IPF.1 Trial data published in 2019 show that some patients with non IPF progressive fibrotic lung disease follow a disease course similar to untreated IPF and that these patients benefit from antifibrotic therapy.2 This finding has shifted research efforts towards finding biomarkers that reliably predict the development of progressive fibroticlung disease with baseline clinical data and imaging data. The evaluation of ILD involves the integration of clinical and imaging data and, in some cases, biological material such as bronchoalveolar lavage, cryobiopsy, or surgical lung biopsy.3 High resolution CT of the chest (HRCT) is routinely used in all patients with suspected ILD and allows non invasive imaging of the disease morphology and the extent of disease. However, visual evaluation ofILD by HRCT is liable to have high rates of inter reader variability and has little sensitivity to changes in disease severity over short follow up periods.4 In contrast, quantitative CT (QCT), which uses computer based techniques to analyse HRCT images, provides an alternative evaluation that is objective and reproducible. Historically, QCT investigations showed that simple measurements based on statistical analyses of CT attenuation values in the lungs can quantify disease severity.5,6 Improvements in computing power have facilitated rapid advances in image analysis, including efforts to use machine learning for the development of more precise and specific digital biomarkers. In this Series paper, we describe the current clinical and research gaps in ILD diagnosis and prognosis and outline how machine learning can be applied to imaging biomarker research in patients with suspected ILD. Clinical and research gaps in ILD For patients with suspected ILD, an accurate, timely diagnosis and a reliable prognostication remain challenging. Incorrect classification of ILD can result in inappropriate treatments, and exposure to risky investigation

FAQs

Why is early detection of interstitial lung disease difficult?

Early ILD is challenging because existing assessment relies on complex clinical, radiological, and pathological processes, and imaging patterns do not consistently reflect disease behavior. High-level visual evaluation can also vary between readers and miss subtle changes early on.

How can machine learning improve prognostication in ILD?

Machine-learning algorithms can use baseline clinical and imaging data to predict the development and extent of progressive lung fibrosis. They can also correlate radiological abnormalities with subsequent lung function decline.

What role does machine learning play in monitoring therapy response?

Machine learning can provide more objective, reproducible quantitative imaging analysis from high-resolution CT, potentially reducing inter-reader variability. It also supports using imaging-derived biomarkers as endpoints in treatment trials.

Aug 4, 2026

Upload by Theodora